美国的能源问题：我们需要一张新电网

The energy grid and electrical grid of the future—it’s not just going to be the dichotomy of generation, transmission, and storage. This next generation of what the grid looks like is going to be much more decentralized.

Why are delivery costs such a big problem?

The grid is aging and brittle. The workforce has aged out. Should we just leapfrog the grid? I need this power now, today.

Today, we have this insatiable thirst for energy. How do you get software almost insidiously on the grid?

The United States needs to get better at megaprojects—things that are a billion dollars, things that are at scale. There is no safety, national defense, or national security without a reliable electrical grid. U.S. energy peaked in 1973 in terms of per capita usage. China’s has increased ninefold over that same time period.

We have some reasons to be optimistic now that things have started to change, or will change further. Why don’t you give some context there? What’s happened, and why should we be excited about what’s coming? Maybe Ryan, why don’t you start?

The history of the grid in the United States was to build big power plants. An industry formed around it, and the grid grew incredibly fast through the 20th century. Then, around the ’80s and ’90s, things started slowing down. Through the early 2000s, the grid effectively froze in the United States.

A big piece of that was that a lot of the energy generation, manufacturing, and heavy industry moved to Asia. For the last 20 years, the grid has effectively ossified. We forgot how to build new power plants. We forgot how to build new power projects and new loads—large data centers, large factories, and large megaprojects.

You say we forgot. Were we not allowed to, or did we actually just lose the skill set?

We were allowed to, but we lost the skill set. You can kind of see it in more extreme examples with nuclear power plants. Although that transition happened decades before, basically the grid itself—the grid operators—forgot how to plan, how to move quickly, and how to do it cheaply. Now we're at this point in time where we are restoring, bringing back manufacturing, bringing back data centers, and there's this highly concentrated demand. It's now "now, now, now" at any price, but they cannot move fast enough. That's what we're seeing today. We talk about data centers and the grid being inflexible to this; it's playing catch-up. We need to do a lot of the growth that happened in China and bring this here and do it incredibly fast.

How did this forgetting happen? And how can this relearning happen, or this learning happen, or kind of retraining happen?

It’s a good question. I think a lot of it is a workforce issue, and I think a lot of it is a policy issue. Historically, the United States started as a bunch of regulated utilities, as a top-down system: big thermal power plants, big transmission lines connecting to substations, and then distribution lines going to individual factories, homes, and things like that.

I think some of the newer technologies don’t necessarily benefit from scale in the same ways that these large thermal plants typically did. This next generation of what the grid looks like is going to be much more decentralized. There’s also an element of relearning what the grid actually is. Is the grid these large power systems and large infrastructure projects, or does it look much more decentralized, where we can eliminate a lot of the wires in between? Delivery costs have increased exponentially. Can we do it in a more dynamic and flexible way?

Solar and batteries don’t need to be massive. You can put them anywhere, including next to the load. This is also something that grid operators are thinking through: How do we do that while also managing frequency, voltage, and things like that without causing the grid to go down? There are a lot of challenges.

If you think about what our grid is, it’s a piece of technology that was designed about 100 years ago, and very little technology on the grid has changed in those 100 years. Why are delivery costs such a big problem? The grid is at capacity. Getting a new project onto the grid today—you sign up for interconnection, and it could take a decade. There’s a backlog of 20-plus years to get a new transformer.

The transformer technology we’re using today is kind of bananas. If you actually look at what makes a transformer, first of all, there’s one company that makes these, and there’s one plant in the U.S. that produces the right type of steel you need to make them. It’s 100-year-old technology, and the waitlist for transformers is insane.

We’re getting to the point where demand is starting to rise again, alongside the calcification of grid technology. Should we just leapfrog the grid? Do we really need to wait in line and wait for this to catch up?

I think there are 2 versions of how you do that. One is: How do you get power generation and power storage as close to demand as possible? That’s a problem for new technology to help solve, because instead of these megaprojects that we’re used to building—well, not used to building anymore—like massive nuclear plants and massive new natural-gas plants, we’re talking about much smaller and more distributed sources of power, bypassing interconnection altogether.

We’re seeing that as a pretty big trend with data centers. Data centers are just building power directly on-site and colocating power with the data center. Microsoft is saying, “I can’t afford to wait 10 years to get an interconnection with the grid. I need this power now, today.” How do you get power more tightly coupled with the load that it’s actually going to serve?

That’s a really interesting problem for technology, also from a software perspective, because if you get generation, storage, and usage all colocated very closely together, that’s a very good problem for AI to solve—reinforcement learning. Stick that in there, and suddenly you get massively efficient systems that you couldn’t get at grid scale.

An interesting point to add on to that is that there’s very little visibility into the grid itself. They understand whether power plants are operating or not, but especially at the distribution level—the power lines you might see outside your home—there’s very little understanding of what’s actually going on there.

There’s a reluctance, especially when you have things like net metering, where I’m sending power from a battery in my home back to the grid. Things get incredibly complicated, and the grid operators don’t have a very good understanding of when they can allow new projects to go online, how much power they can allow, or when to actually cut people off.

There are a lot of these policies. Interconnection is the general term, but states like Texas have a much more lenient policy: You can build wherever you want, but if we need to cut it off from the grid, we’re going to do so. It’s a connect-and-manage approach.

Other states conduct incredibly long feasibility studies across a variety of scenarios. The entire grid is at peak capacity, but they want to make sure that this specific project can stay online 24/7. That ends up creating massive delays. There are a lot of policy approaches here as well.

There are a bunch of technologies called grid-enhancing technologies. An average power line might be used at 50% capacity, but it needs to be designed for peak capacity—for the summer, when everyone has their AC on. There are a lot of sensors and other technologies that could be placed there, so you have a much more dynamic view of what our infrastructure actually looks like.

When we have these new technologies, we can use the infrastructure we have much more efficiently.

David, what are your reactions to this conversation so far? Where are some areas you’re particularly excited about, or reasons to be optimistic?

I think the reason we’re having this conversation is that we’re touching on a bunch of topical themes. We’re at a moment in time where, exactly as Ryan said, the grid is aging and brittle. The workforce has aged out. We had to go out and hire and train entire specialized crews—specialized people who work with cement and concrete, and specialized people who work with steel—to build the large Vogtle reactors in Georgia.

We put them on Vogtle reactors 3 and 4. We turned them on—a huge win. Then those people went back to building highways, bridges, or something else. Instead of putting them on Vogtle 5, 6, 7, 8, 9, and 10 and building a massive crescendo of nuclear power, we put those people back into the general workforce.

And so we just are not learning our lesson there on the workforce. At the same time, we have this insatiable thirst for energy, whether it's EVs, data center compute for AI, or just generally a shift toward more and more consumption of electricity, or even just the reshoring and manufacturing and all these things that are very, very electron-heavy.

I think at the same time, there's a piece that we hardly ever talk about, which is resiliency and not having people be as dependent on the interconnectedness of the grid. For those of us who are in the tech world, we talk about distributed compute and how important it is to have distributed compute and have networks be able to suffer and survive through segmentation and things. But the grid is very interdependent. Even in the U.S., there are really only a few major regions that can segment themselves off.

But when you deploy solar, batteries, an SMR reactor, or your own power generation on-site for your own data center, you don't have to worry about how brittle the grid is because you're fairly resilient to it. I think that's a component of the energy grid and electrical grid of the future. It's not just going to be the dichotomy of generation, transmission, and storage. As Erin brought up, you might do all 3 of those things in the same place and not have to worry about how robust the grid is or how capable the grid operators are.

I think that's a dimension that was never important to people before, but it's important today. You can certainly imagine that if you're the military, you care about having reliable access to power at all your forward operating bases and even at your home military bases. You just cannot lose your ability to have electricity.

I think all these things are coming together at once, and it's a really exciting moment in time. I think it's buoyed by the fact that we're also at this sort of technology inflection point where AI can help some of these things not just be a consumption driver, but even be an enabler in facilitating more efficient use of electricity, better monitoring of the grid, and better ways to go through the regulatory and permitting process, which is onerous in many cases.

Yeah, I think building on that, Texas has famously had massive grid failures. Several years ago, when a big heat wave came through, the grid couldn't keep up with all the air conditioners that were running, and people saw massive power outages. Everyone was really mad. People were saying, "It doesn't work. Deregulation doesn't work."

What has Texas done in the couple of years since that happened? They have absolutely flooded the grid with solar capacity. Texas has doubled its solar capacity in the last approximately 3 years, and with that they've deployed thousands of batteries. One of our portfolio companies, Base Power, is one of the players here, but there are many battery companies deploying across Texas to provide storage for that solar power.

If you look at the performance of the Texas grid versus the performance of the New York and surrounding-area grid during this heat wave, I must have seen 10 news articles this morning about how well the Texas grid has done. There's an elasticity and ability to react to changes—to very quick changes—in demand without having to change baseload power. You can't build a new natural gas plant or a new nuclear reactor overnight, but solar is just so insanely cheap. It's basically having a giant, massive nuclear reactor in the sky that will go forever.

Texas isn't a green state. This isn't a political issue. But why aren't we deploying the world's cheapest form of power literally everywhere we possibly can and then just putting batteries everywhere? There should be batteries everywhere. It's kind of bananas to me that batteries as a topic have recently gotten caught in the political crosshairs. As a society, we really just need to be good at power storage and batteries. This shouldn't be a controversial topic.

We invented the lithium-ion battery, and yet today, if you want to buy a battery—whether it's for a drone, the grid, your car, or whatever it is—you're either buying a battery made in a lights-out factory in China, or you're buying a battery produced in Vietnam by a Chinese company. There's no meaningful effort in the U.S. to change that.

This is a really critical problem, not just to manage power load on the grid, but for powering all of the things we need to power the next generation of innovation in the United States. I think we'd be hard-pressed on the American Dynamism team to think of a company we've met with that had interesting technology in the last 2 years that doesn't have a battery in it somewhere.

As a country, we need to be investing in battery technology and battery manufacturing. By the way, if China decides that whatever your company is doing that uses batteries doesn't align with what they like, or they want to punish you, being cut off from the ability to buy batteries from China is incredibly punitive to a company. We've certainly seen that happen with some of our startups.

And then you find out quickly that the ability to procure and source batteries from places that are not in China is very, very difficult. If you extrapolate that out to what would happen to our whole country if we were just unable to buy batteries from China, it could be catastrophic in a very, very short period of time.

Just to add a quick point on the grid side of batteries: If the rest of the country, which is presumably watching what's going on in ERCOT, the grid operator in Texas, sees that Texas can prove you can deploy these decentralized, distributed energy resources to flatten these peaks, provide more resiliency, and ultimately lower the price of electricity, then every state should go and do this.

There's a very complex web of deregulated and regulated entities when it comes to the grid. Of course, there are a lot of different policy, workforce, and political reasons why not everywhere is this decentralized world. It will probably be more complex than just these deregulated energy-only markets that Texas works with.

I think the United States needs to move incredibly fast to make this happen and hook up batteries and solar panels, making it easier and cheaper to do so. Even collocation for large loads is still a very politically fraught issue. Utilities are pushing against this, and it's still really hard to hook up solar and batteries to your home.

I think it's actually cheaper to put residential solar on your home in Germany than in the United States. That's largely a permitting and installation issue. That's crazy. That should not be the case.

Erin, I believe the quote in your college yearbook was "Drill, baby, drill." How do you think about your love for oil and gas alongside other sources of energy?

My parents will be shaking their heads if they hear this. Broadly speaking, our approach to energy in the U.S. just needs to be "yes, and." You look at the atrophy of our power buildout over the last 30, 50, or whatever you name your time frame years compared to, let's say, China. If we want to accomplish the goals that we've set out as a society to accomplish over the next decade, we need more power. It's a matter of "yes, and."

I think solar and batteries are extremely important, but there is a place for oil and gas. I cut my teeth at Palantir working in oil and gas. My husband worked in oil and gas. The first check I wrote at A16Z isn't an oil and gas company. This isn't me coming with a particular agenda around carbon. I'm realizing that we basically need every tool in our toolkit, and we should be using technology to deploy whatever makes the most sense, wherever it makes the most sense, at scale.

If we're talking about the energy mix—where we're at today and where we think we're headed—if I were to make a personal bet, it's that solar and batteries are going to continue to be incredibly cheap and deploy incredibly fast.

Spin up and spin down.

Yeah, and I think that's already the case, in a way, but I think that will continue. To be very clear, you need all different types of energy. You're going to need true baseload, dispatchable power. It's going to be gas, it's going to be nuclear, it's going to be geothermal, and there's going to be a lot of hydro as well.

As you attach more of these renewable resources, or these non-reliable resources—which are incredibly cheap and work most of the time—this sort of long-tail risk, once you get to 50% to 75% of the grid, is going to become very, very expensive. You need a lot more battery backup and things like that. I think it's going to be very complex and different for many different regions, but it certainly isn't going to be all of any given resource.

When you look at the changing nature of load over the next decade, some fraction of that is going to come from data centers. I would say it's probably overstated how much data centers contribute to the growing load in the United States over the next decade.

Data centers generally are baseload. If you're training a model, you're largely using a dedicated amount of power for the long term. Maybe there are some fluctuations if you're doing more inference, but I would generally say data centers represent baseload. You also have things like electric vehicles, heat pumps, and air conditioners. You have industrial autonomy, which may or may not be running 24/7.

So you're going to have some increase in the base level of power we as a society need, while continuing to increase the size of the peaks and troughs of how we use energy on a day-to-day basis. We should be thinking about designing our grid, our energy mix, and our power sources around what those loads look like, and not oversolve for either baseload or variable power.

I think, just to put this in more tangible terms, the peak summer load in places like California might be half of what it is in winter, or something like that. It depends on what climate you're in. The concept of baseload is: do you build all the plants you'd need for the 100 gigawatts of power you're going to need in the winter, when in the summer, half the year, you're only going to need 50? So what would be baseload?

What you need to do in modern civilization is make sure that every time you turn on the switch, the power is working. How you actually match supply with that very fluctuating load, both daily and seasonally, is very complex. Today, you might have to build a natural-gas peaker plant that might only operate for a week a year.

That's an incredibly expensive asset that is only going to be delivering very expensive power, but it is only needed when all the other resources are tapped and it's that last couple of megawatts of power. The alternative now is demand response or batteries on the grid. You can say, "Instead of building this $10,000-per-megawatt-hour plant, I can just make it so everyone's thermostat in this area turns down a couple of degrees." In aggregate, that means I don't need to build that large asset or pay that expensive premium. There are a lot of things like that.

I do think, though—pushing back on that—that the American consumer will flatly reject that level of dictation over how they use their thermostat. I think a more likely outcome is that you can do it on the compute side and just say, "Look, these 3 racks of the data center are just going to go offline during the peak summer heat when you're running your AC."

This is not a critical job.

Right? It's a noncritical job. It's not a mission-critical job. It's a back-office job, and you're just going to run it at night instead of during the day. You're going to pay less for electricity for that benefit.

I do think that as we build out—I don't know if I agree with Erin that AI is not going to suck up all the compute. I think compute is going to—I think Constellation just turned up a new nuclear reactor or is reactivating a reactor, and I think Meta immediately sucked up all of the power that they're going to generate, or 9/10 of it or something, from the new Constellation reactor that Meta signed the contract extension for. I think we actually probably are underestimating the amount of compute that we're going to soak up with electricity over the next 10, 20, 30, 40, 50 years. The amount of data we're going to start storing—I mean, the amount of video we create per minute has just ballooned way beyond anyone's expectations. I'm sure the same will be true for AI compute.

I think once you start getting into robotics and autonomy, if you think about compute expansively, I totally agree.

Yeah, but those things are going to be much more responsive than, "Do I want to have my room be 74° instead of 71°?" Anyone who's done business in Tokyo in the summer knows that, as a nation—by the way, Japan has done this—it is absolutely terrible.

It's horrible. We're not going to do that in America. But I will say, do that in America. You're on the 40th floor of a Japanese building, wearing a suit—by the way, because you have to wear a suit—and it's swelteringly hot. Everyone is walking around as if they're not miserable, but they are miserable, and you want to jump out a window, but the window won't open. It is one of the worst things.

Why doesn't France work in the summer? This is why.

It's terrible. Exactly.

So we hate this idea. I'm spoiled living in California.

I will say one of the biggest proponents of this, or current users, is crypto mining. These are flexible, demand-responsive loads, right? Those operations will just turn off their compute when it's cost-effective.

But that's important for the grid. You can build these assets and have the demand for power here—they're going to soak up that demand—but if it gets far too expensive, they will also shed that demand.

But in the U.S., my guess is that this is going to—I mean, this already is reflected in the fact that you have peak-load pricing. For me, my power is 10 times more expensive between 4:00 and 9:00 p.m., so we don't run the washing machine. I leave it to organizations or individuals to figure out how to manage that, but just charge people more for power.

It's a little bit of a non sequitur, but I like that we keep talking about oil and gas, natural gas, batteries, solar, and nuclear. We talked about—Ryan even mentioned hydro, which of course is totally viable in some places. The thing that nobody ever brings up anymore, except, I think, a very fringe group, is wind power. I'm very happy to hear that nobody here is stumping for wind.

I think wind is incredibly cheap when it's working. It's just that you don't know if it's going to work. You kind of know solar is going to work on a reliable schedule: the sun's going to be out, and it's going to be working, sparing some cloudy days, but there's still always something coming through. But the wind might not blow for a week.

I think it's worse than that. I think I read that, globally, 1/3 of all wind turbines are out of service at any given time. The other thing is, I think wind is the only power-generation mechanism where, when you get too much of the input, the blades of a wind turbine feather and turn off. Whereas if you get too much sun for—

Solar, you stop. You just max out.

Too much water and hydro, that's not a problem. But too much wind and the wind generator turns off. It's just—who wants a system where you get more of the input you want and then it stops working?

It's also just extremely hard, dangerous, and specialized to service. You see those videos of people climbing the ladder up to the top of the wind turbine.

Yeah. It's like—

Yeah.

Grid operators look at wind: it's great when it's working, but they can't plan for it. They have to build other capacity to supersede that if it's not going to be there when they need it.

So fine. I see wind. We can move past wind.

Yeah, I agree. So no wind. But I do think demand response—going back to the point that Ryan brought up—monitoring the grid is really important, and being able to send signaling on the grid is really important.

You have to remember that we're all used to the internet, which has bidirectional communication and messaging, a data layer and control layers, a full control plane, and things like that. The electrical grid doesn't really have that. To be able to send messages and monitor the grid itself without an overlay network is very hard. A lot of people now just do it out of band using the internet. That's one of the challenges that people are starting to address.

Yeah, it's wild how much of a mystery what's happening on the grid is at any given time. We really have very little visibility, and it's very hard—

—for, I think, centralized utilities to deploy meaningful software to understand that. So when we think about, as VCs, what types of things we look at and what we get excited about, I think companies that are approaching this monitoring from the opposite direction are interesting.

How do you get software almost insidiously on the grid? How do you start learning more about demand and generation as close to the source as possible, and then try to feed that information back to each other? The idea that you're going to go sell a software tool to a PG&E or similar and have a reasonably speedy top-down implementation, where you actually get good signal and metrics and can do interesting things with that data, seems to me a little bit unbelievable.

Something very interesting that I learned is that a lot of load forecasting is basically the task of determining when plants need to go online. There’s usually a 24-hour-ahead, or day-ahead, market that basically says, “You need to run your plant at this time.” Then there’s this sort of supply-and-demand match to a price, and there’s a merit order. It’s complex, but that’s sort of how it’s done.

Most of this load forecast is done by looking at the weather. One of the best inputs is where the homes are, how many people are there, and what the temperature is going to be. That often ends up being the largest factor that goes into the modeling.

But now, if we have all these connected resources—solar, EV chargers, and all of this stuff spitting off data, telemetry, and things like that—we’re going to get a much better look at how load is actually being forecast. That’s going to become real-time, which is going to help a lot with understanding where we actually need to build and what the actual price of power is. Then you can start making these markets a lot more efficient.

When you look at energy desks for the big hedge funds or energy-trading companies, their weather person is usually the highest-paid person on the desk besides the portfolio manager. Those climate and weather PhDs working on trading desks are absolutely raking it in because they’re kind of like gods right now—there’s very little other data.

That’s why, when you see what goes on in Texas with heat waves and things like that, if they get it wrong by even a couple of degrees—if they think it’s going to be sort of hot, but it actually gets really hot—that’s when you get these crises that end up causing a lot of strain on the grid. Then you have to turn on all these very expensive plants, and you get the headlines that are usually also the worst for the environment.

Yep. Do you want to let us know if you have any reactions to this, or otherwise give us the state of nuclear? Where are we right now? What are the major bottlenecks? What are we excited about?

The biggest thing that’s shifted in the last 3 or 4 years in nuclear is that everybody now acknowledges that nuclear energy is clean energy. I think that’s been a major shift in public sentiment and perception. Nonetheless, there are still major political headwinds with nuclear that need to be overcome.

Taiwan, for instance, turned off its last nuclear reactor. This is unbelievable. This is an island country that is 7 days away from a total blackout if it gets an oil-and-gas blockade from China, so it can’t bring in ships to deliver oil and fuel. At any given time, they’re 7 days away from a total blackout, and they turned off their last nuclear reactors.

Why did they do that? Because they caved to a political minority, I would say.

A very loud, vocal—

Environmental activist reasons.

Yeah, this party ran on a commitment to turn off the reactor before they realized how stupid it was. It’s just colossally stupid, by the way. Turning off a reactor—a real, full-scale reactor—isn’t like an SMR, where you can just flip it on a few days later or a month later. With these large reactors, it can take years to turn them back on. So that was just terrible.

Broadly, I think the tailwinds for nuclear are just getting stronger as people recognize that it is clean energy. I think there’s still messaging work to be done. We should stop calling spent fuel nuclear waste because it’s really not waste. Almost all of it can be recycled and reused. People do need to recognize that those tailwinds are shifting.

I think people understand that it’s baseload power, right? It’s not dependent on only working during the daytime. It’s not like hydro, where you have to be around an appropriately configured water source.

One of the largest inhibitors to creating new power plants in this country is not that we can’t do it—we can. There’s a huge regulatory-permitting morass that has to be swum through. It’s incredibly expensive, requires an army of consultants, many tens of millions of dollars, and many thousands of pages of applications, documentation, and process review.

Again, this has to do with building the power plant, getting the fuel, transporting the fuel, and storing the fuel. Each step along the way is extremely laden with regulation and policy. Some of that is for good reason, but finding ways to better navigate that and make it more efficient is an area that a lot of companies and people are focused on now.

Actually, I think the government is also focused on how to streamline the approval process for a new reactor and how to start approving new reactor designs.

The last thing I’d say is that right now, if you’re going to put a lot of energy and work into building a nuclear power plant, you want to build a really big one. We largely only see really big power plants in this country, like the AP1000s that we turned on in Georgia. Those, again, came in, I think, 10 years late and multiple billions over budget.

We are now starting to see movement from the government—from the Department of Defense, the Department of Energy, and the national labs—to really try to create a more fast-track process for small modular reactors, or even microreactors, that use a much safer form of fuel, use much less nuclear fuel, and use a different kind of nuclear fuel that’s not nearly as risk-prone as the kind of nuclear fuel or nuclear material you’d use in a weapon. It’s not nearly as enriched to the same degree, and it’s not even the same material. That process is now gaining a lot of steam.

We have an investment in a company called Radiant Nuclear. They’re building a factory that creates what is effectively an SMR. They would probably call it a microreactor. It’s a 1-megawatt reactor. It can be put on the back of an 18-wheeler and shipped around. You can move it to where you need power.

If there’s been a natural disaster, like a hurricane, and you need to bring in power overnight, you could bring in a few trucks with 4 or 8 of these reactors and power up a whole city after a disaster. That kind of flexibility and power is really compelling.

The last thing I’ll say about the SMR and microreactor side is that the United States military spends an incredible amount of money just dealing with the logistics of moving fuel around to forward operating bases. Anytime we do a military exercise, anytime we’re engaged in a conflict, the movement of fuel factors in as a primary concern and consideration.

We’ve read reports that they spend well over $200 a gallon, sometimes up to $400 a gallon, for diesel, effectively to get diesel into the right place at the right time. You can just imagine that having a nuclear reactor you can put on the back of a C-130 and fly around the world to wherever you need power, drop it in the middle of the desert, turn it on, and have power for 5 years is an incredibly compelling value proposition.

There’s no question that nuclear needs to be part of the equation. Not only is that baseload power, but on the SMR and microreactor side, it gives us incredible flexibility and grid resilience. There should not be a single military base in this country that’s not nuclear-backed from a power standpoint, because if the grid goes down—whether it’s from a cyberattack, instability, demand issues, or cascading failures—you want to be able to fail over to nuclear power and not worry about the runway lights turning off.

Yeah. Especially as we start to look at the electrification of our weapon systems, our military vehicles, and our drones, all of those need to get charged up somewhere. How better to charge them than with a nuclear reactor?

The other thing I’ll add to your nuclear comment is that I think the advantage of nuclear—and I think Radiant has done very well in really leaning into this—is the power-density factor. If you want a reactor that is reliable and power-dense, you want it to operate at very high temperatures, and you want as highly enriched fuel as you possibly can where it makes commercial sense. So you want HALEU fuel, and you want to serve customers that will pay the premium for that. They’ll buy this reactor because they know it’s going to work.

If you’re doing that, you want to have economies of scale on the manufacturing side. You want it to be done out the door and not need to be assembled on-site. You don’t want to have constant maintenance.

I think the other sorts of reactors that we see, maybe on the civilian side, have a problem. If you build a reactor in a factory, or build modular components in a factory, but you still need to do construction work on-site, you’re still a construction company.

Even if the technology is there, I would argue that a lot of the existing AP1000 technology is quite good. Other countries can do it quite cheaply. China is using a very similar design. The UAE just built one incredibly cheaply.

They have very similar nuclear regulatory frameworks, and obviously their regulatory bodies might move faster and things like that, but they're not completely ignorant of some of the concerns. So I think, at a high level—and maybe this is a much broader question—the United States needs to get better at megaprojects: things that are billion-dollar projects, things that are at scale. I would argue that the NRC is a big component of why it's expensive, but I think it's also the same reason that it takes a billion dollars to build a bike lane in San Francisco.

Why don't we have high-speed rail in California?

Yep.

We might not have high-speed rail in California because nobody wants it, and nobody wants $100 to LA all the time.

Well, nobody—sorry. Nobody wants it where they're building it.

Sure. Yeah.

Bakersfield is not a prime destination.

It could be.

I want a train from San Francisco to LA that takes an hour and a half.

Me too.

Yeah. And we debate a lot internally about where it makes sense for VCs and VC capital to plug in. Arguably, we're not going to move the needle on these multibillion-dollar megaprojects in the US. We're probably not the best people to figure out how to capitalize and build a multibillion-dollar project in California to generate power for the grid.

But I do think there's a role for technology at every single layer and every single phase of how megaprojects get built. How do you use AI to navigate site selection? How do you use tools to move through the various permitting processes faster? How do you use AI to help you do extremely complex and interdependent project management better and more effectively?

You have a project with 4,000 people working on it, everyone engaging with different suppliers and timelines that are dependent on each other. How do you get all those things to align better so that you don't get these 10-year delays, so that projects actually happen on time and on budget and, as a result, attract private capital backers? I think there's a role for technology here. What that looks like—

We've seen a lot of companies that maybe 5 years ago were primarily trying to sell to utilities and grid operators, which is incredibly painful, incredibly difficult, and perhaps rightfully so. They have poles in the ground that are 50 years old. Why would they trust a 2-year-old company to sell them software? Are they going to be around in 20 years? This is a fair question to ask, especially for something as critical as the grid.

But now they're developing this software, and there's such demand for understanding how grid operators might think and potentially getting there faster or having different conclusions. Now you can go to data centers, people who want to build solar farms, or people who want to build massive battery farms, and you can sell a very small—

Or even individual people who want to make sure that their power isn't going to go out and that they're not going to be caught without energy during an important moment in their lives.

Yeah. Now everyone cares. There's a lot more money from people who care about what the grid is actually going to think and where they can build—where there is excess capacity. Maybe I'm connected to the grid, but I also need some battery and solar backup, or a Radiant microreactor or something like that, to be used in certain situations.

It's a lot more complex, this sort of microgrid setup, but it's the way we're headed, and software is going to be a big piece of that.

I want to hear more about our requests for startups, or things that we want to exist that we haven't yet discussed. Or, maybe put differently, I'm curious where we think there's the most bang for the buck in terms of the issues we've been talking about. If there were a regulatory intervention or some sort of technological unlock, what comes to mind?

One area where there's probably a venture-scale software company to be built is around grid-management monitoring infrastructure. We see this in the IT landscape; we see it in the OT landscape, but we don't really see it in the grid. There is no Splunk for the electrical grid. There's no Palo Alto Networks for the electrical grid yet.

There's a whole bunch of things that mirror the IT and OT landscape, whether it's around cyber, monitoring, logging, and analytics. There's no Looker for the electrical grid yet. None of these companies exist. I'm not sure if it's 3 separate companies. I'm not sure if it's 1 company, but there is a big company to be built in really managing and monitoring the grid, helping to orchestrate and even deal with some of the things Ryan spoke about around demand response: coordinating that, creating those marketplaces, and tracking all those incentives.

I think when we see a company that we think can really be the breakout company there, we would lean into it. I also think around project planning and development: how do you make it faster and easier to build projects within the current regulatory framework? How do you do site selection? How do you navigate permitting? How do you navigate project management? How do you navigate your construction supply chain?

We're starting to see companies pick off pieces of that, but broadly speaking, there's room for tech and software in that kind of project development space as well.

Anything that can bring generation capacity or storage capacity closer to load, I think, is going to be very compelling. A lot of the time, it's less about novel technology, but it's sort of a system integration or an innovative business model.

I think something like radically changing that experience is—and I implore everybody to go home and check their power bill—they'll now often separate the delivery cost from the actual generation costs. What we've seen, and we've sort of mentioned it, is that the cost to generate electricity—the cost of power—has dropped immensely: gas, solar, things like that. But the cost to actually deliver that electricity has increased a ton. In net, it's sort of not changed.

I think that's terrible, and I think we all agree that's bad. I think there's a lot of opportunity in bringing that generation capacity closer to load. In some ways, this is sort of a more liberalizing force. We all should have our own backup. We all should have our own technology. I think there are a lot of really interesting ways to do that and scale it.

Overall, as the grid gets more heterogeneous, all of the seams and intersections between things create so much more opportunity for technology than when you had a single utility managing a single source of power centrally, distributed out broadly.

All the regulation and permitting and policy frameworks that we have in this country can be thought of as part of the infrastructure that we all have to live and work with and interact with. So companies that really facilitate that—and I would say applying AI to navigating the permitting process. Nuclear is a good example again: a nuclear reactor application, a fuel transport license, or a fuel manufacturing license.

These things have thousands and thousands of pages of regulation and documentation that go with them. You make one small change in your application, and it has these reverberating effects. You have to update all your documents elsewhere. If you're the regulator trying to go through all these applications, it's just incredibly onerous. It's borderline impossible to imagine that a regulator can even possibly get it right. You could argue that it's actually not possible; they just make a best effort.

But AI could actually help with these things. It could help the applicants go through the process of filling out their applications and saying, "Hey, this is where you should drill down. This is where you should clarify." It can look at all previous published applications and say, "This is how you need to tailor it." You can probably make it 85% the same, and then, based on your design or your location or whatever, make some modifications.

Then the regulator can do the same and say, "Look, here's an application that came in. Highlight all the areas where we need to drill down or make sure everything is in order, or show me the things that are different from every other nuclear fuel transport application we've ever seen. Are they using the rail infrastructure? Are they using the national highway infrastructure to move the fuel?"

AI can just automate all these things that take armies of consultants months or years to do. They can be brought down to minutes or hours.

So I think there’s potentially—I’m not sure how big of a company—but I think potentially there’s a very large company to be built there.

If our check sizes were in the billions, not just the millions, which we’re in, Jason Harwood, you never know. How would our strategy change?

Ryan, you’re talking about megaprojects. What kinds of megaprojects?

I think you need even more than just billions. It’s tens of billions, hundreds of billions. It’s such a tough question. I think there’s tons of policy around this as well.

I hesitate to say we should look to how China has built up its grid, but I think the elephant in the room is that in the early 2000s, they were experiencing blackouts. This was a very common thing, and it was horrific. But now they’ve, I think, 4×’d their grid in the last couple of decades.

And the way they’ve done this is by basically deploying generation capacity, building hydro, building massive storage facilities, and, of course, producing tons of batteries through BYD and CATL. They’ve built HVDC—these large, high-voltage transmission lines. I would do all of that.

I mean, I would look to all of it, and whether or not it’s a good investment is going to depend on a number of factors. But it is much more the infrastructure projects—the glue that connects the stuff together. I think our lens today is looking at technologies that enable a lot of this more flexible grid, but I think there are also going to be these large infrastructure projects, the webbing in between it. And I think software is a big piece of it that we’re spending a lot of time looking at.

How is it going to be connected to the rest of the grid? How are we going to move this power around if it’s really sunny in the Southwest and solar is going to be really cheap? Is there an efficient way to move that to New York or something like that? China’s done this effectively. And I think if you had hundreds of billions of dollars spent—or trillions of dollars—what does the grid look like? It’s going to be a lot more interconnected.

Another answer to your question, or a different answer to your question, is that I think the energy industry is probably, in the medium to long term, one of the prime spots to deploy physical autonomy.

When you think about applications of robotics, whether they’re humanoids or more task-specific robotics, I think these are often dangerous jobs. We’re talking about manufacturing jobs to build up small modular reactors or batteries or whatever. I don’t know what the shape of the company is here, or how reliant it would be on some of the robot-learning work that’s happening, but I do think that as we scale our energy capacity, there’s going to be a pretty massive application of industrial robotics to the energy sector.

You know, we survived the greatest nuclear disaster in U.S. history just recently, when we finished the Vogtle 3 and 4 reactors and let all those employees go back to other jobs.

So I think if we were writing a billion-dollar check into power, what we would do is just give jobs to those people and not let them go back to whatever they were doing before they were building nuclear reactors. We would really work to streamline the process to make sure that we go build Vogtle 5, 6, 7, 8, 9, and 10 in all these different states around the country, and just put these people to work for the next decade-plus building reactors. I don’t think it’s particularly our opportunity, but I do think it’s an opportunity for somebody to do.

Labor broadly—I mean, this is a little tangential—but when Microsoft was building its new data center in Georgia last year, at one point they had, on staff at Microsoft or on contract, more than a third of the electricians in the state of Georgia. They basically maxed out: they hired every single electrician they possibly could.

So I don’t think it applies just to electricians. To your point, it’s the cement mixers, mechanical engineers, and nuclear engineers. How do we actually train the next-generation energy workforce that we’re going to need to modernize the grid?

By the way, these are very high-paying jobs where you don’t have to check your email on your phone at 9:00 p.m. after you go home from work. They’re high-paying, good-exercise jobs, and they’re relatively low-stress. These are good jobs for people.

I think one more comment on this, and it’s more of an industrial-policy question. We’re talking about specific things, but oftentimes that just moves the bottleneck. We could solve a lot of the grid-connection hookup and build a lot of transmission lines, but then we need more transformers. To build more transformers, you need more electrical steel, and you can do the same sort of equation for much of the supply chain.

Batteries are another good example: okay, great, we’re building cells, but then we also need active materials, and we need to mine and things like that. It’s a whole effort of examining our infrastructure and our supply chains, and you need to do all of it. I think that’s a complicated question and an expensive question.

People underestimate how critical and important a resilient, reliable, dispatchable electrical grid is to our national security. You cannot have national defense and national security without reliable electricity. It’s just not possible. All these things we’re talking about are about the upside—capitalizing on AI compute, the transition to electric vehicles, and our insatiable thirst for electricity. But at a fundamental level, there is no safety, no national defense, and no national security without a reliable electrical grid.

People want cheap, reliable, and clean power, in that order. I think that’s largely how we should think about our energy policy, and I think that’s the direction we’re going. We need to make sure we stay aligned with that.